Is storage is more important that scan read in your requirement , Then yes you can perfom compression on it because LZ4 is the best in my opinion interms of compression / performance when compared with snappy or zlib.
whats the cardinality on the string column ? low or high ?
If the cardinality is not high I would vote for dictionary encoding without compression. The dict encoding will save you a lot of space. However I dont know the internals of Kudu, where is the limit, how much distinct value can be in the dictionary.